CN-121812055-B - Chicken coccidiosis resistance space-time drift deduction method and system

CN121812055BCN 121812055 BCN121812055 BCN 121812055BCN-121812055-B

Abstract

The invention relates to the technical field of drug resistance monitoring and treatment, and discloses a chicken coccidium drug resistance space-time drift deduction method and a system, wherein the method comprises the following steps of 1, receiving a multi-mode heterogeneous data set; the method comprises the steps of 2, constructing a phenotype feature vector according to production performance recovery hysteresis quantity, recovery acceleration extreme value and fecal spectral entropy, 3, inputting the phenotype feature vector into a conditional reversible neural network to output a drug-resistant genotype frequency distribution vector, 4, constructing a student model at a cloud end by taking the trained conditional reversible neural network as a teacher model, inputting a drug-resistant genotype frequency distribution vector matrix and reputation weight into the student model to output a global drug-resistant feature map, 5, constructing an asymmetric evolutionary game model to generate a drug-resistant time-space evolutionary thermodynamic diagram, and 6, inputting the regional average drug-resistant genotype frequency distribution into a reinforcement learning model to output a sequential drug-use scheme.

Inventors

CAI HAIMING
ZHU YIBIN
ZHANG XIAOHUI
ZHANG JIANFEI
SUN MINGFEI
LIAO SHENQUAN
Qi nanshan
LI JUAN
LV MINNA
LIN XUHUI
Song yongle
CHEN XIANGJIE

Assignees

广东省农业科学院动物卫生研究所

Dates

Publication Date: 20260508
Application Date: 20260312

Claims (8)

1. The chicken coccidium drug resistance space-time drift deduction method is characterized by comprising the following steps of: Step 1, receiving dynamic growth data, pathology characterization data, drug intervention data and environment covariates of a chicken flock after drug intervention to form a multi-mode heterogeneous data set; Step 2, obtaining production performance recovery hysteresis quantity based on a multi-mode heterogeneous data set by adopting a recovery hysteresis calculation formula, obtaining a recovery acceleration extremum by adopting a recovery acceleration calculation formula, obtaining fecal spectral entropy by adopting a Fourier transform formula, and constructing a phenotype feature vector according to the production performance recovery hysteresis quantity, the recovery acceleration extremum and the fecal spectral entropy; Step 3, constructing a conditional reversible neural network, inputting the phenotype characteristic vector into the trained conditional reversible neural network, outputting a drug-resistant genotype frequency distribution vector, and generating a local drug-resistant report according to the drug-resistant genotype frequency distribution vector; Step 4, based on a federal knowledge distillation method, constructing a student model at a cloud end by taking a trained conditional reversible neural network as a teacher model, inputting a preset public anchor point data set into the teacher model, outputting a drug-resistant genotype frequency distribution vector matrix and a digital signature by the teacher model and uploading the drug-resistant genotype frequency distribution vector matrix and the digital signature to nodes of a federal blockchain, verifying the digital signature by the nodes of the federal blockchain through intelligent contracts and calculating reputation weights, inputting the drug-resistant genotype frequency distribution vector matrix and the reputation weights into the student model, and outputting a global drug resistance characteristic map by the student model; Step 5, constructing an asymmetric evolution game model based on the global drug resistance characteristic map to obtain a reaction-diffusion equation set, and generating a drug resistance time-space evolution thermodynamic diagram in a plurality of future periods by solving the reaction-diffusion equation set; Step 6, based on drug resistance time-space evolution thermodynamic diagram, obtaining regional average drug resistance gene frequency distribution by adopting a thermodynamic diagram space aggregation algorithm, inputting the regional average drug resistance gene frequency distribution, a candidate drug set and a culture period into a reinforcement learning model, and outputting a sequential drug use scheme; the step 5 comprises the following sub-steps: Step D1, constructing an asymmetric evolutionary game model based on a global drug resistance characteristic map, and defining a strategy set of prevention and control as chemical synthesis drugs, polyether ionophore antibiotics and vaccines, wherein the strategy set of a pathogenesis side is a sensitive strain and a drug resistance mutant strain; Step D2, defining a payment matrix, and calculating the fitness of each drug-resistant genotype of the coccidium under different drug pressures by adopting a payment fitness function; Step D3, constructing a reaction-diffusion equation set based on the fitness of each drug-resistant genotype of the coccidium under different drug pressures; Step D4, solving a reaction-diffusion equation set through a finite difference method to generate a drug resistance time-space evolution thermodynamic diagram in a plurality of future periods; the payment fitness function is: ; Wherein, the Is the first The adaptability of the drug-resistant genotype of the coccidian, Is the first Probability of use of the seed drug in the current region, Is of drug-resistant genotype For medicines Is used for the survival rate of the tolerance of the strain, To maintain drug-resistant genotype The cost of the consumed physiological energy, D, is the total number of the drug classes currently examined; The set of reaction-diffusion equations is: ; Wherein, the Is the first The genotype is spatially located And time of day Is used for the frequency of (a), Is the first The adaptability of the genotype of the seed, For the average fitness of the population, In order to spatially expand the tensor, Is the divergence of the diffuse flux.
2. The method for deriving chicken coccidiosis resistance spatiotemporal drift as claimed in claim 1, wherein said step 2 comprises the sub-steps of: A1, constructing an actual observation growth curve based on dynamic growth data, and constructing a standard infection-free growth curve based on ideal growth data of healthy chicken flocks under the condition of no coccidium infection; a2, constructing a response residual function based on the actual observation growth curve and the standard infection-free growth curve, wherein the response residual function is as follows: , wherein, Is a standard infection-free growth curve, For actually observing the growth curve, t is a time variable; a3, calculating initial production performance recovery hysteresis quantity by adopting a recovery hysteresis calculation formula based on a response residual function, and carrying out normalization processing on the initial production performance recovery hysteresis quantity to obtain production performance recovery hysteresis quantity; A4, obtaining an initial recovered acceleration extremum by adopting a recovered acceleration calculation formula based on a response residual error function, and carrying out normalization processing on the initial recovered acceleration extremum to obtain a recovered acceleration extremum; a5, obtaining an initial fecal spectral entropy based on pathological characterization data by adopting a Fourier transform formula, and carrying out normalization processing on the initial fecal spectral entropy to obtain fecal spectral entropy; And A6, constructing a phenotype characteristic vector according to the production performance recovery hysteresis quantity, the recovery acceleration extremum and the faeces spectral entropy.
3. The chicken coccidiosis resistance spatiotemporal drift deduction method according to claim 2, wherein the recovery hysteresis calculation formula is: ; wherein, I is the production performance recovery hysteresis, For the time of the intervention of the medicine, In order to view the window time of the window, At time instant for responding to residual function Is used as a reference to the value of (a), Is a time decay factor; the recovery acceleration calculation formula is as follows: ; wherein V is the initial recovered acceleration extreme value, For responding to residual functions Is a second time derivative of (2); The fourier transform formula is: ; Wherein H is the spectral entropy of the feces, Is the first in the gray level histogram of the stool spectral image Normalized frequency of the individual gray levels.
4. The method for deriving the drug resistance of chicken coccidiosis by time-space drift according to claim 1, wherein in the step 3, each farm is respectively constructed with a respective conditional reversible neural network; The conditional reversible neural network of each farm is optimally trained through the private historical data of the farm and a biological constraint loss function, wherein the biological constraint loss function is as follows: ; Wherein, the Is a negative log-likelihood of being a function of the number of bits, For the L1 sparsity constraint, In order to cross the constraints of resistance to drugs, 、 Are weight coefficients.
5. The method according to claim 1, wherein in the step 4, the specific process of verifying the digital signature by the node of the federal blockchain through the intelligent contract and calculating the reputation weight comprises the following sub-steps: Step C1, after a node of the federal blockchain receives a drug-resistant genotype frequency distribution vector matrix and a digital signature, an intelligent contract verifies the validity of the digital signature through a public key cryptography mechanism; when the digital signature is valid, calculating a plurality of cosine similarities between the drug-resistant genotype frequency distribution vector matrix of the node and the drug-resistant genotype frequency distribution vector matrix of other nodes, and then calculating an arithmetic average value among the cosine similarities to obtain average similarity; Step C3, when the average similarity of the nodes is smaller than a preset threshold value, rejecting the frequency distribution vector matrix of the drug resistance genotype of the nodes; The reputation weight formula is: ; Wherein, the Is the first The reputation weight of the individual node(s), Is the first The degree of historical contribution of the individual nodes, Is the first The data quality score of the individual nodes, 、 Are all the weight coefficients of the two-dimensional space model, Is the first The degree of historical contribution of the individual nodes, Is the first Data quality scores for individual nodes, N being the total number of nodes.
6. The method for deriving chicken coccidiosis resistance time-space drift according to claim 1, wherein in step 4, the training mode of the student model is to update the global parameter by minimizing the KL divergence of the student model output and all weighted teacher model outputs, and the specific formula is: ; Wherein, the As a result of the new global parameter(s), Is the first The reputation weight of the individual node(s), For the KL divergence, the average value of the power supply is calculated, Is the first A matrix of frequency distribution vectors of the drug-resistant genotypes of the individual nodes, The frequency distribution vector of the drug-resistant genotype is output for the student model, As a set of data for a common anchor point, Is a parameter of the cloud global student model.
7. The method for deriving chicken coccidiosis resistance spatiotemporal drift as claimed in claim 1, wherein said step 6 comprises the sub-steps of: step E1, obtaining the frequency distribution of the regional average drug resistance genes by adopting a thermodynamic diagram space aggregation algorithm based on drug resistance space-time evolution thermodynamic diagrams, wherein the thermodynamic diagram space aggregation algorithm is as follows: ; Wherein, the Is genotype At a time step Is used for the frequency of (a), For the number of spatial grids, For the time-space evolution thermodynamic diagram of drug resistance, For the stock weight of the farm l, Coordinates of a v-th grid node; e2, constructing a full-period optimization objective function based on the regional average drug resistance gene frequency distribution, and performing optimization training on the reinforcement learning model by using the full-period optimization objective function, wherein the optimization objective function is as follows: ; Wherein, the For the total duration of the cultivation period, As a function of the economic loss, For the moment of time Is a region average drug resistance gene frequency distribution after polymerization, For the moment of time Is used for the administration strategy vector of (1), As a function of the level of drug resistance accumulation, And Are all weight coefficients; and E3, inputting the regional average drug resistance gene frequency distribution, the candidate drug set and the culture period into the reinforcement learning model, and outputting a drug administration strategy.
8. A chicken coccidiosis resistance spatiotemporal drift deduction system, characterized by being used for realizing the chicken coccidiosis resistance spatiotemporal drift deduction method according to any one of claims 1-7, comprising the following units: The multi-mode data receiving unit is used for receiving dynamic growth data, pathology characterization data, drug intervention data and environment covariates of the chicken flocks after drug intervention to form a multi-mode heterogeneous data set; The phenotype feature vector construction unit is used for obtaining production performance recovery hysteresis quantity based on the multi-modal heterogeneous data set by adopting a recovery hysteresis calculation formula, obtaining a recovery acceleration extremum by adopting a recovery acceleration calculation formula, obtaining fecal spectral entropy by adopting a Fourier transform formula, and constructing a phenotype feature vector according to the production performance recovery hysteresis quantity, the recovery acceleration extremum and the fecal spectral entropy; The inverse dynamics calculation unit is used for constructing a conditional reversible neural network, inputting the phenotype characteristic vector into the trained conditional reversible neural network, outputting a drug resistance genotype frequency distribution vector, and generating a local drug resistance report according to the drug resistance genotype frequency distribution vector; The federal knowledge distillation coordination unit is used for constructing a student model at the cloud based on a trained conditional reversible neural network serving as a teacher model based on a federal knowledge distillation method, inputting a preset public anchor point data set into the teacher model, outputting a drug-resistant genotype frequency distribution vector matrix and a digital signature by the teacher model and uploading the drug-resistant genotype frequency distribution vector matrix and the digital signature to nodes of a federal blockchain, verifying the digital signature by the nodes of the federal blockchain through intelligent contracts and calculating reputation weights, inputting the drug-resistant genotype frequency distribution vector matrix and the reputation weights into the student model, and outputting a global drug-resistant characteristic map by the student model; the drug resistance drift deduction unit is used for constructing an asymmetric evolution game model based on the global drug resistance characteristic map to obtain a reaction-diffusion equation set, and generating a drug resistance time-space evolution thermodynamic diagram in a plurality of future periods by solving the reaction-diffusion equation set; the sequential decision generating unit is used for obtaining the regional average drug resistance gene frequency distribution based on the drug resistance time-space evolution thermodynamic diagram by adopting a thermodynamic diagram space aggregation algorithm, inputting the regional average drug resistance gene frequency distribution, the candidate drug set and the culture period into the reinforcement learning model, and outputting a sequential drug use scheme.

Description

Chicken coccidiosis resistance space-time drift deduction method and system Technical Field The application belongs to the technical field of drug resistance monitoring and treatment, and particularly relates to a chicken coccidium drug resistance space-time drift deduction method and a chicken coccidium drug resistance space-time drift deduction system. Background Chicken coccidiosis (Coccidiosis) is an intestinal parasitic disease caused by Eimeria (Eimeria) protozoa that severely damages the poultry industry. At present, the prevention and control of the disease mainly depend on polyether ionophore antibiotics (ionophores) and chemical synthetic drugs (chemical drugs). However, coccidian Resistance to existing drugs is increasingly severe due to long-term, high-strength and often non-standardized medication in intensive farming, and exhibits complex multi-Drug Resistance and cross-Drug Resistance characteristics. The prior art has the following remarkable pain points in the aspects of coccidium drug resistance monitoring and treatment. First, the monitoring means are slow and expensive, and traditional drug resistance detection relies on "chicken body assays" (In vivo) or large-scale PCR gene sequencing. The former has long period (a plurality of weeks of isolated feeding), high cost and strict ethical examination, while the latter has high cost, and cannot be popularized in common commercial farms, so that drug resistance data is often delayed from epidemic outbreaks. Secondly, the data island and privacy barrier problems are prominent, and the drug resistance data is closely related to the drug application program, the management level and the survival rate of the farm, and belongs to highly sensitive commercial confidentiality. Data often do not communicate with each other among the culture groups, even among different branches of the same group, so that a medicine resistance time-space evolution map of the whole domain cannot be constructed, and regional joint defense joint control is difficult to implement. In addition, lacking dynamic evolution prediction capability, existing studies focus on "current" static drug resistance levels, lacking dynamic prediction of "future" drug resistance drift. The change of the genetic frequency of coccidian population under drug pressure is a dynamic process conforming to evolutionary game theory, and the prior art lacks a mathematical deduction model based on a 'drug-pathogen' countermeasure view angle, so that a drug administration strategy is always passive. Therefore, the application solves the technical problem of dynamically predicting the drug resistance time-space drift trend and outputting the drug use scheme on the premise of not revealing the privacy of the original culture data. Disclosure of Invention The application mainly aims to provide a chicken coccidium drug resistance space-time drift deduction method, which realizes dynamic early warning of coccidium drug resistance without relying on large-scale gene sequencing by constructing a decentralized phenotype-genotype federal reverse mapping network under the premise of strictly protecting commercial privacy, thereby reducing cost and improving speed. And constructing a medicine-coccidium dynamic countermeasure model by combining the evolutionary game theory, simulating adaptive drift paths of coccidium genotypes under different regional medication pressure strategies, and generating a global medicine resistance time-space evolution thermodynamic diagram. And finally, outputting a medication scheme through the reinforcement learning model. Meanwhile, a chicken coccidiosis resistance space-time drift deduction system is also provided. In order to achieve the above purpose, the present application adopts the following technical scheme: a chicken coccidian drug resistance space-time drift deduction method comprises the following steps: Step 1, receiving dynamic growth data, pathology characterization data, drug intervention data and environment covariates of a chicken flock after drug intervention to form a multi-mode heterogeneous data set; Step 2, obtaining production performance recovery hysteresis quantity based on a multi-mode heterogeneous data set by adopting a recovery hysteresis calculation formula, obtaining a recovery acceleration extremum by adopting a recovery acceleration calculation formula, obtaining fecal spectral entropy by adopting a Fourier transform formula, and constructing a phenotype feature vector according to the production performance recovery hysteresis quantity, the recovery acceleration extremum and the fecal spectral entropy; Step 3, constructing a conditional reversible neural network, inputting the phenotype characteristic vector into the trained conditional reversible neural network, outputting a drug-resistant genotype frequency distribution vector, and generating a local drug-resistant report according to the drug-resistant genotype frequency distribution vector; Ste